Agentic

The content reuse agent: think once, publish ten times

Last Thursday, you spent forty minutes explaining to a client why his implementation had stalled. You gave him a distinction that no one else in your market makes quite so clearly. The client understood straight away. What has happened with that since then, outside that one room?

Last Thursday you spent forty minutes showing a prospect's CFO why per-seat pricing punishes his growth and usage-based pricing rewards it. Nobody in your category draws that line so sharply. He nodded and asked for a proposal. Where does that explanation live now, outside that one Zoom call?

On Tuesday you spent forty minutes showing a shipper why his Eindhoven–Bremen lane loses money. The rate per trip ignores the return leg, so the empty kilometres come out of your margin. Nobody in your region lays it out that sharply. He nodded. Where does that explanation live now, outside that one canteen?

Last week you sat in a steering committee and explained to a client's CFO why his fixed-price migration was running over. You drew a sharp line between a change request and an assumption that never made it into the statement of work. He got it at once. Where does that explanation live now, outside one set of minutes?

Last week you sat across from the technical director of a German customer. In half an hour you explained why a retrofit of his fifteen-year-old line would pay back more than a new machine. He nodded and asked for a quote. What has happened to that reasoning since, outside that meeting room in Stuttgart?

I want to run the voice test on my own source

30 mins · senior consultant · no slide deck

The problem isn’t the writing

Almost everyone misdiagnoses this. You think you don’t have enough time to write. That’s not true.

Almost every founder misreads this one. You think you need to hire another content marketer. That is not the problem.

Almost every haulier says the same thing: between planning and the driver shortage there is no time left to write. That is not the problem.

Almost every agency looks for the cause in the wrong place. You think content dies for lack of non-billable hours. That is wrong.

Almost every machine builder misreads this. You think your engineers have no time to write alongside live orders. That is not true.

The outsourced work has become the cheapest part of the process. McKinsey reported a saving of one and a half million hours within its own organisation in 2025, and those hours were mainly spent on research, synthesis and initial drafts. That is precisely the stage that you currently carry out manually.

McKinsey, 2025

The real sticking point lies ahead

The real bottleneck lies before that. Ideas are never documented. They arise during a client meeting, a demo or a discussion with your CTO, and then they disappear. This continues to be the case because documentation has no deadline and no designated owner, whereas publishing does.

The bottleneck sits earlier. Your best thinking never gets captured. It happens in a QBR with your largest account, in a Slack thread with your CTO about the roadmap, in the appendix of your board deck. Then it vanishes. Capturing sits in no OKR; the next release does.

The bottleneck sits earlier. Your best thinking happens in the annual rate negotiation, at the morning briefing in planning, or when you and your fleet manager rework the Maut surcharge. Then it is gone. Capturing it has no deadline and no owner. An invoice has both.

The bottleneck sits earlier. Your team thinks out loud in a pre-sales call, a sprint review or the post-mortem after an outage, and afterwards it is written down nowhere. Capturing has no ticket and no owner. Billable work has both, so it wins every time.

The real bottleneck sits earlier. The thinking is never captured. It happens in the post-calculation review, on the phone with a service engineer on site, at the engineering kick-off, and then it is gone. Capturing has no deadline and no owner. The next order has both.

A glass wall with the sea beyond it, light streaming through the glass
Work it out using your own figures

Work it out using your own figures. Take the number of hours you spend each month rewriting and reformatting existing material. Multiply that by your own full-rate hourly rate. Divide that by the number of pieces you actually publish that month.

The denominator is therefore the number of articles published per month, not the number of hours. Here’s a sample calculation, purely by way of illustration: six hours multiplied by ninety euros equals 540 euros; divided by four published articles, that works out at 135 euros per article.

So the denominator is published pieces per month, not hours. Worked example, purely illustrative: eight hours of your product marketing manager, at seventy-five euros, is €600, divided by three published pieces, is €200 per piece.

So divide by the number of published pieces, not by the hours. An illustration, not a measurement: you and your sales manager put eight hours a month into content, at 85 euros that is 680 euros, divided by two pieces that actually went live, is 340 euros per piece.

So the denominator is published pieces per month, not hours. Worked example, purely for illustration: eight hours of your solution architect, at €85 cost per hour, is €680, divided by two published pieces, is €340 per piece. And those were hours you could have billed.

So the denominator is the number of pieces published per month, not the number of hours. A worked example, purely as illustration: eight hours from your head of engineering, at 95 euros, is 760 euros, divided by two published application stories, is 380 euros per piece.

Enter your own two figures and you’ll have your cost price per publication.

Second-order costs are more expensive

Second-order costs are higher. By this stage, your buyer is asking their first question of a model, not your sales page. A model can only cite what already exists. Your competitor, who has half your knowledge but does publish weekly, is on the shortlist – whilst you are not.

The second-order cost weighs more. The head of operations at a German mid-sized firm asks ChatGPT which tool fits his DACH rollout long before he books your demo. A model can only cite what exists. The competitor with a thinner product but a comparison piece every week makes his shortlist. You do not.

The second-order cost weighs more. A shipper's supply chain manager now asks a model first which 3PL can handle ADR storage and daily lanes into the Ruhr. A model only cites what is online. The carrier with half your fleet who writes every week makes that shortlist. You do not.

The second-order cost is higher. The IT manager looking for a partner for his cloud migration asks a model first, before any tender goes out. A model can only cite what exists online. The agency with half your experience that publishes a technical case every week makes the longlist. You are not on it.

The second-order cost is higher. A German buyer first asks a model which special-machine builder can deliver a packaging line with format changes under ten minutes. A model can only cite what exists. The competitor with half your experience, who publishes an application story every month, makes that longlist. You are missing.

[ Four questions ]

Four questions you can answer this evening

  1. How many hours a month do you spend rewriting something you’ve already explained once?

  2. How many sources lasting more than ten minutes did you record last month, whether as text, a recording or a transcript?

  3. Based on the figure above, what is your cost per published piece, and would you be willing to pay that price again for the next piece?

  4. If you were to place the last three items from your company next to those of your biggest competitor and cover up the logos, would anyone on your team be able to tell which ones are yours?

    Put your last three release posts or customer stories next to your biggest competitor's and cover the logos. Could anyone on your sales team point out which ones are yours?

    Put your last three LinkedIn posts about a new lane or warehouse next to your biggest competitor's and cover the logos. Could your planner tell which ones are yours?

    Put your last three case studies or tech blogs next to your biggest competitor's and cover the logos. Could your account manager spot yours, or does every one say 'on time and on budget'?

    Put your latest trade-fair news, product brochure and LinkedIn post about a handover next to your biggest competitor's and tape over the logos. Can your sales manager tell which are yours?

How this agent works, and where you remain in the loop

The loop is the PRAL loop from this pillar.

Perceive

The agent reads a resource that you provide – an article, a webinar recording or a transcript of a real customer conversation.

The agent reads a source you supply: a recorded discovery call, a churn exit interview or the memo behind your last pricing change.

The agent reads a source you supply: the recording of a quarterly review with a shipper, a damage claim file or your notes on a new rate quote.

The agent reads a source you supply: a product demo recording, the post-mortem after an outage or an architecture decision record from your wiki.

The agent reads a source you supply: a factory acceptance test report, the notes on a change request or the recording of a quotation meeting.

Reason

He extracts the claims, examples and opinions and determines which derived forms they take.

Act

He submits the derived documents to a review queue – namely, posts, an email summary and an FAQ entry.

Learn

Your corrections to tone and wording form the stylistic memory that guides the next round.

L2, AI-Augmented

To be clear: this is L2, AI-Augmented, in the Capgemini classification. The agent makes a suggestion; you validate it. It is not L3. Capgemini estimated in 2025 that around fifteen per cent of processes would run at L3 or higher, and publishing under your name does not fall into that category. Publishing is outward-facing and practically irreversible, so it falls into the tier where a human must give prior approval. Anything in which a customer is identifiable must be subject to approval.

Capgemini, 2025

Where you get stuck in a loop

You remain at two points in the loop, and those two are not interchangeable. During recording, because the agent cannot create a source that does not exist. And during the voice check just before publication.

What a line manager does in week one

Appoint one person to be responsible for recording. Choose one source per week. Schedule a 30-minute review session in the diary. Set out the retention period for recordings in writing.

Make your head of product marketing own capture. Pick one recorded sales call a week. Book thirty minutes of Friday review with the founder. Set the retention period for recordings in writing.

Make your sales manager the owner of capture. Pick one conversation with a shipper each week. Block thirty minutes of review on Friday after weekly planning. Put in writing how long you keep recordings.

Make your delivery manager owner of capture. Take one sprint review a week as the source. Book thirty minutes of review with your lead developer. Put in writing how long client-code recordings are kept.

Make your head of service the owner of capturing. Pick one live order as the source each week. Book thirty minutes of review after the order meeting. Write down how long customer recordings are kept.

Count how many times you rewrite a draft to a greater extent than just a light edit. That figure is your baseline, and without a baseline, you won’t know in a quarter’s time whether it made any difference.

A close-up of the key, with the beard clearly in focus against a dark background
[ By sector ]

The same workflow, three different diagnoses

The difference between sectors does not lie in the words. It lies in what you measure, how you record it and at what time.

SaaS scale-up

  1. 01 Measurement unit: one demo call per sprint in which a prospect raises an objection that has not yet been addressed anywhere in your materials.
  2. 02 Tool: the transcript of that recording, with the three minutes during which the answer was given highlighted.
  3. 03 Release: the product owner reviews the roadmap to ensure it is factually accurate, as a commitment to a feature constitutes a promise to the customer.
  4. 04 Clock: sprint rhythm, two weeks. Measure, over three sprints, how often the same objection still had to be answered in real time.

Logistics

  1. 01 Unit of measurement: one anomaly that occurred for the second time this month. Not a conversation, but a disruption.
  2. 02 Tool: a five-minute voice memo, recorded by the planner on the day itself, plus the line from the ticketing system.
  3. 03 Release: the operations manager checks whether the case can be traced back to a single shipper and anonymises it if necessary.
  4. 04 Clock: monthly, and linked to the peak period. You record data during the busy month and carry out your assessment during the quiet month.

Mechanical Engineering

  1. 01 Unit of measurement: one commissioning or service report per quarter. The thinking behind this is already set out in writing, but nobody outside the company reads it.
  2. 02 Document: the report itself, with the customer’s name, location and serial number removed, plus the technician’s notes.
  3. 03 Approval: an engineer signs off on every technical claim. An incorrect figure relating to a tolerance is not a typographical error but a product claim.
  4. 04 Timeline: per project milestone, with a quarterly review. Your project’s lead time is the timeline, not the content calendar.
A sailing boat on the open water with a billowing sail, heading towards the horizon
Please bring

Try it out this evening, before you buy anything

First, the data rules, as these apply before you paste anything in. This workflow involves less personal data than lead qualification or accounts receivable management: you enter a single source of your own choosing, not a database. ‘Lighter’ does not mean ‘zero’. A recording of a customer conversation constitutes personal data for everyone who took part in it.

Before you use this

You need to have this sorted before you use a real recording.

  • Prior consent from all participants.
  • A data processing agreement with your model supplier.
  • A business subscription, not a consumer account.
  • To know in which region the processing takes place.
  • A written retention period for recordings and transcripts.
  • A documented escalation process with a name attached to it.

Our rule of thumb is to keep things for ninety days.

Our estimate; not a measured standard.

Regarding your colleagues, to put it bluntly: yes, this records what an employee says during a conversation and processes that information. That is why the recording is of the employee. They give their consent in advance, they review the draft before publication, and they can withdraw it without giving a reason. The recordings are not used for appraisal or performance assessment. Make this clear before the first recording, not afterwards. In a documented procurement committee, the works council has a right of veto here, and rightly so.

For the test below: use an anonymised source, or one of your own articles. The disqualification rule is set out in the instructions themselves: if there are fewer than three, the source is too thin to be reused.

Je bent redacteur. Ik plak hieronder twee teksten. [BRON] is mijn eigen ruwe bron:
een transcript, een artikel of een notitie. [AFGELEIDE] is een stuk dat daaruit is
afgeleid. Werk in drie stappen en gebruik uitsluitend wat in [BRON] staat.

1. Bronsterkte. Noem de concrete claims, getallen, voorbeelden en meningen in [BRON]
   die niet in een willekeurig artikel over dit onderwerp zouden staan. Geef het
   aantal. Polariteit: een hoger aantal is beter.
   Regel: bij minder dan drie is de bron te dun om te hergebruiken.

2. Generiekheidstest. Citeer elke zin uit [AFGELEIDE] die je net zo goed over een
   willekeurig ander bedrijf in deze markt zou kunnen schrijven. Geef het aantal.
   Polariteit: een hoger aantal is slechter.

3. Herkomst. Noem elke bewering in [AFGELEIDE] die niet terug te voeren is op [BRON].
   Citeer per bewering de letterlijke zin uit [BRON], of schrijf "niet in de bron".

Sluit af met een oordeel van een zin: publiceren, herschrijven, of eerst een betere
bron opnemen. Geef geen herschrijving, alleen het oordeel en de drie lijsten.

[BRON]:
[AFGELEIDE]:

Apply this to your most recently published piece. If the source strength falls below three, using this agent is a quicker way of saying nothing.

Run this on your latest launch post. If it reads like any other changelog and the source scores below three, this agent buys you a faster way to say nothing.

Run this on your latest post, say the one about your new cold store. If all it says is 'flexible and reliable', this agent buys you a faster way to say nothing.

Run this on your latest case study. If it only says 'agile, scalable, happy client' and scores below three on source strength, you are buying a faster way to say nothing.

Run this on the press release about your latest handover. If it only says ‘high-quality bespoke’ and scores below three, this agent buys you a faster way to say nothing.

Where this officer doesn’t belong

If you don’t have a source, this solves nothing. In that case, RENEW is your starting point, not an agent. Nor does it form part of an original position: a new idea does not arise from reformulating an old one. It does not apply to regulated statements – such as medical claims, safety specifications or financial commitments – unless an authorised person approves every sentence. And it does not apply to you if, for you, writing is the very act of thinking itself. Automating that away will cost you exactly what you wanted to multiply.

Read more

The voice test on your source.

I’d like to compare one of your sources with one of your published pieces and show you what the voice test does with them.

I want to lay one of your recorded sales calls next to your current pricing page and show you what the voice test makes of it.

I will put the transcript of your last rate negotiation next to your latest newsletter and show you what the voice test makes of it.

I would like to put one post-mortem or pre-sales recording from your team next to your latest case study and show you what the voice test does.

I would like to put one of your service reports next to your latest reference project and show you what the voice test does with it.

I want to run the voice test on my own source

30 mins · senior consultant · no slide deck